While antenna simulations can be an accurate predictor of gain, and duplication of known antennas can produce clones with predictable gain, these methods do not actually measure antenna gain. There are methods of measuring fields close to an RF excited around an antenna, called “near field scanning”, that simulate and predict the gain of an antenna.  

However, all true measurements of antenna gain come back to the Friis Transmission Equation. I have written the equation here in the form of path loss, with units of decibels.

It says that the path loss between two antennas is the sum of their gains plus a term that relates frequency and their separation distance. In any test setup, you would know the test frequency and antenna separation distance, thus the sum of the antenna gains involved is easy to calculate.

Path loss is something that is easy to measure with a VNA, or any source receiver. You are probably already familiar (and equipped) to measure two port loss. Once you subtract the frequency/separation term from your measured path loss, you have the sum of two antenna gains. If one antenna gain is known, then the unknown is resolved. If both antennas are identical, you may assume that each contributes half of the total sum game. A more detailed article on measuring gain can be found here.  

In actual antenna gain testing, there are many other practical considerations, such as reflections, sweeping frequencies, and directional rotation (patterning). Some helpful educational links are available here.